Technical Term Recognition with Semi-supervised Learning Using Hierarchical Bayesian Language Models

Abstract

To recognize technical term, term dictionaries or tagged corpora are required, but it will take much cost to compile them. Moreover, the terms may have several representations and new terms may be developed, which complicates the problem further, that is, a simple dictionary building can’t solve the problem. In this research, to reduce the cost of creating dictionaries, we aimed at building a system that learns to recognize terminology from small tagged corpus using semi-supervised learning. We solved the problem by combining a tag level language model and a character level language model based on HPYLM.

We performed experiments on recognition of biomedical terms. In supervised learning, we achived 65% F-measure which is 8% points behind the best existing system that utilizes many domain specific heuristics. In semi-supervised learning, we could keep the accuracy against reduction of supervised data better than exisiting methods.